Exponentially Weighted Dynamic Kernel Principal Component Analysis Algorithm and Its Application in Fault Diagnosis

Abstract

Kernel principal component analysis(KPCA) can use kernel function to solve nonlinear problem,and it has excellent nonlinear approximation ability,but traditional KPCA cannot deal with dynamic problems.A new method is proposed on the basis of exponentially weighted dynamic kernel principal component analysis algorithm,a multi-variable weighted autoregressive statistic kernel principal component model is built,Q statistics are selected to judge whether the system has fault or not,the concrete calculation steps of fault diagnosis are given.The new method is tested on the hydraulic pump,the end-cover vibration signal is processed by using wavelet packet,the fault feature vector composed of 13 time and time-frequency domain features is extracted.Test results show that the new method can renew the principal component model and control limit Qa,rationally utilize real-time dynamic information,better deal with dynamic problem,and through calculation and comparison,can select appropriate weighted factor and obtain good effect of fault diagnosis,so this method is feasible and effective.